Various methods have been used in the past to predict or test for the failure performance of semiconductor devices. Most manufacturers of such devices use these various methods to prevent weak devices from failing when used by the customer. The manufacturer desires good performance from the device in order to build device reliability, prevent returns, reduce service calls by the manufacturer, and prevent customer dissatisfaction.
A well known phenomenon typically associated with the life of semiconductor devices is known as "infant mortality." This term is used to describe the relatively high failure rate observed in new components. The initial high failure rate of components normally drops to a low value after a short period of operation, and continues to remain low for a relatively long period of time. In order to avoid the effects of "infant mortality" in semiconductor devices, it is common to "burn-in" such components until the period of infant mortality is past. This "burn-in" typically involves alternately passing a high forward current through the device, and applying a reverse bias voltage approximately equal to the rated peak inverse voltage or temperature changes in the device. Examples of these "burn-in" systems and methods may be seen in U.S. Pat. No. 5,047,711 by Smith et al. entitled "Wafer-Level Burn-In Testing of Integrated Circuits", U.S. Pat. No. 5,030,905 by Figal entitled "Below a Minute Burn-In", and U.S. Pat. No. 4,215,309 by Frey entitled "High Efficiency Switching Circuit."
"Burn-in" testing, however, fails to detect various problems associated with the semiconductor devices that may not be detected by the application of forward current alone and also problems that may occur when the device is beyond the initial high failure rate point in the degradation curve. As an example, high voltage surges often occur in electrical systems that expose the semiconductor devices to higher voltages than specified by the manufacturers. The instantaneous power surge that occurs typically determines the breakdown voltage wherein device failure occurs. The breakdown voltage may be tested for or predicted by a system or method such as seen in U.S. Pat. No. 4,307,342 by Peterson entitled "Method and Apparatus For Testing Electronic Devices."
Other testing systems and methods have been attempted for testing radiation being emitted from semiconductor devices such as light emitting diodes ("LEDs") or laser diodes. Some of these systems and methods may be seen in U.S. Pat. No. 4,578,641 by Tiedje entitled "Systems For Measuring Carrier Lifetime of Semiconductor Wafers", U.S. Pat. No. 4,797,609 by Yane entitled "LED Monitoring Without External Light Detection", U.S. Pat. No. 5,065,007 by Tanaka entitled "Apparatus For Measuring Light Output From Semiconductor Light Emitting Element", and U.S. Pat. No. 4,611,116 by Batt entitled "Light Emitting Diode Intensity Tester."
Also, testing systems and methods have been attempted where the semiconductor devices such as P-N junction diodes or bipolar transistors are pulsed with a high current surge to induce damage to the semiconductor junction of interest. Examples of these systems and methods may be seen in U.S. Pat. No. 3,978,405 by Petree entitled "Damage Thresholds Of P-N Junction Devices By A Current Pulse Method" and U.S. Pat. No. 4,301,403 by Hawkes et al. entitled "Electrical Circuit Testing."
Another known method of testing semiconductor devices is by statistical sampling a group or lot of devices developed from the same material and then projecting the performance of the entire group of devices based on the statistical sample. For example, in more mature LED material, such as GaAs, GaAsP, or AlGaAS, a wafer containing LED dice is typically scribed and sawn, and then statistically sampled. The LED samples are initially tested for various optical and electrical performance characteristics. Based on the degradation or other testing of the statistical LED samples, the entire wafer of LED dice is either accepted or rejected. A description of this technique for LED semiconductor devices may be seen in the book "Light Emitting Diodes: An Introduction" by K. Gillessen and W. Schairer, .sctn. 3-3-7 at 98-99 (1991). Although helpful for mature semiconductor material where performance is predictable, such statistical sampling fails to be a proper technique to select less mature semiconductor material such as silicon carbide (SIC).
Other techniques such as probing a wafer of semiconductor devices, such as LEDs, are also known. These techniques typically involve "wafer probers" which are instruments used to test the quality and yield of completely processed wafers. The prober puts a probe on the contact of a first device on the edge of a wafer and allows the user to perform various measurements. After the measurements on the first device have been finished, the prober moves the wafer a controlled distance to the next diode. The data obtained during a measurement cycle is then evaluated in an accept-reject manner and bad diodes marked by a dot of ink. A description of this technique may also be seen in the book "Light Emitting Diodes: An Introduction" by K. Gillessen and W. Schairer, .sctn. 3-3-6 at 97 (1991).
The wafer probing technique, however, like the statistical sampling, may be appropriate for some mature semiconductor materials, but may not always provide a method for measuring performance of less mature semiconductor material over time, thereby causing the customer dissatisfaction problems previously described. Also, the performance of devices manufactured with this less mature material may not be statistically predictable because of its relatively recent development. For example, the use of SiC for various semiconductor devices, including light emitting diodes, has only recently become a commercially viable alternative. Devices made from these materials may be seen in U.S. Pat. Nos. 4,918,497 and 5,027,168 by Edmond, one of the inventors of the present application, entitled "Blue Light Emitting Diode Formed In Silicon Carbide". Hence, these less mature materials require a different testing apparatus and method.
Therefore, there is still a need for an apparatus and method for testing semiconductor devices wherein performance is not statistically predictable for a group of devices and which predicts performance over time.